„Privilegien“ bei der Erforschung maschinellen Lernens

Research exemptions aim to place scientists in a privileged position by fully or partly excluding scientific research from the scope of a regulatory regime or by providing more favourable regulatory conditions. Scrutiny usually arises when an exemption appears too narrow to sufficiently protect scientific freedom. Remarkably, even the full exclusion of scientific research from the scope of a regulation can appear as insufficient in this regard. Such seems to be the case with the research exemption in the Artificial Intelligence Act. By only excluding research conducted for the »sole purpose« of science, the regulation draws a distinction between such research and other, more »product-oriented« forms of research. Considering some particularities of the scientific field of machine learning - especially the strong emphasis on the transfer of research results into practice and the so-called reproducibility crisis - a de facto pressure to comply with the regulation may arise. The effects of such pressure are ambivalent and raise a number of questions requiring further research.